Why keep folding after the timeout expires?

[arisu]

If a WU is finished after the timeout expires but before the deadline passes then the folder will get the base level of points with no QRB. As soon as the timeout expires, the work servers will send out a duplicate WU to someone else. But why let the first folder keep folding instead of giving them a new WU? What is the benefit of receiving duplicate WU results?

Or to put it another way, if someone has passed the timeout and they know someone else has gotten the WU they are still working on, would it be more beneficial for science for them to dump that WU and get a new one, or to keep folding and return a duplicate WU?

[muziqaz]

No benefit. Please, raise a request on fahclient-bastet github to implement the functionality for client to monitor the deadlines and delete the WU which went beyond it. Also good idea would be to notify the user that certain device could not finish WU in time

[arisu]

So getting rid of the timeout/deadline dichotomy in the client would be a good idea, and immediately request a new WU once the timeout expires? Basically treating the timeout as the deadline, and getting rid of the deadline?

[muziqaz]

Ah, crap I misread the post. Sorry
Well, the system was set up this way to not discourage people with older hardware from folding, and also encourage faster hardware usage at the same time. And the reason why a WU is reassigned again while first one is still folding is probably because of not to hold the project back too much.
As you participate more, you will see there are many concepts which seem to be a waste of scientific resources, unfortunately working out a near perfect system is near impossible. There will always be some waste, heck look at the current GPU fahcore chaos. One would think that is a massive detriment to the project (which it is), and one would think that should be addressed half a year ago and not now, but reality is that FAH project is volunteer heavy. People do their life stuff first and then on their spare time try to help out here. There is no one in this project who have 100% dedication of their time towards FAH researchers have their projects to care for, project lead has his Uni responsibilities and personal life, project dev has his hands full with FAH infrastructure and client development. GPU fahcore developer is a student, who has his own stuff to deal with. The rest of us are volunteers.
And, btw, perfect design for some, might be discouragement to participate for others

[arisu]

Oh I see! So there's no technical benefit for them to keep folding, but seeing points awarded for their scientific (non-)contribution, despite wasting resources, might encourage them to upgrade to better hardware? And that may ultimately be better for the project than simply raising the bar to kick out people with devices that only meet the deadline but not the timeout?

The core22 issue is a mess, agreed. It (and all similar issues) could be completely alleviated if they statically-link the cores. But at least the immediately-dumped WUs get reassigned right away. Better than if they got to 99% before dumping.

[calxalot]

A reassigned WU may not be finished. So it can still be beneficial for folding to continue past timeout.

[muziqaz]

So if a person contributes to the project with hardware which doesn't meet timeout, but meats deadline, they feel like they are contributing (positivity is the key). If we rename timeout to deadline and dump old hardware, that will be massive negative impact towards all those users. Project itself has been pushing for elitist hardware for a while. This time out thing is last stand holding people who do not upgrade every release cycle, but when they do, they are still with the project, instead of forgetting about it, if they were excluded completely at the time when they were with slower hardware

Now that I think about it, timeout/deadline system was introduced simply to entice newer hardware usage more than anything.
Also, quite a lot of times, slower hardware is more stable, there might be a case where slow hardware folding on a WU past timeout, that WU is assigned to some gamer kid who thinks their whitebox system is 100% stable (while it is not), they crash the WU, or something else happens, and before you know it, the first folder with slower hardware actually returned their WU before deadline, and they actually are the first to return it

When I mentioned gpu fahcores, I meant all GPU fahcores have identical issues
core22 does not work on modern distros, core 23 has issues with random older and random modern distros, core 24 has issues with old distros.
There is glibc version mismatch issue plaguing all cores
core26/27 is meant to fix that